Behavioral Bandits : Analyzing the Exploration Versus Exploitation Trade-Off in the Lab

This paper uses a laboratory experiment to analyze how individuals resolve an exploration versus exploitation trade-off. The experiment implements a single-agent exponential bandit model. We find that, as predicted, subjects respond to changes in the prior belief, safe action, and discount factor. However, we commonly find that subjects give up on exploration earlier than predicted. We estimate a structural model that allows for risk aversion, base rate neglect/conservatism, and probability mis-weighting. We find support for risk aversion, conservatism, and probability mis-weighting as potential factors that influence subject behavior. Risk aversion appears to explain the finding that subjects explore less than predicted